Professional Documents
Culture Documents
Energy
journal homepage: www.elsevier.com/locate/energy
a r t i c l e i n f o a b s t r a c t
Article history: The relationships between consumption, energy use and greenhouse gas (GHG) emissions from Swedish
Received 5 September 2012 households are analysed using both cross-sectional multivariate regressions and a decomposition
Received in revised form analysis of GHG emissions between 1993 and 2006 into underlying trends. The analyses cover 104
26 April 2013
consumption categories and include both direct and indirect energy use. The results from the cross-
Accepted 6 January 2014
sectional analysis confirm previous results from other countries showing that total expenditures is by
Available online 31 January 2014
far the most important determinant of households’ energy use and GHG emissions with expenditure
elasticities of 0.77 and 0.85, respectively. Households living in single-family houses were also found to
Keywords:
Greenhouse gas emissions
cause higher total energy demand than households living in apartments. Age also showed a positive
Households relationship with both total energy use and emissions, while the level of education had a low signifi-
Determinants cance. The decomposition analysis showed that improvements in technical energy efficiency over time
Consumption patterns had twice as large impact on energy use as the change in composition of consumption. When the effect
Energy efficiency of energy efficiency improvements was deducted from the energy trend, the relationship between total
expenditures and energy use was found to be similar to the cross-sectional relationship.
Ó 2014 Elsevier Ltd. All rights reserved.
0360-5442/$ e see front matter Ó 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.energy.2014.01.019
J. Nässén / Energy 66 (2014) 98e106 99
socio-economic parameters (total expenditures, household size, the effect of structure was larger than that of changing sectoral
education, age) and physical structures. Such determinants should energy intensities. Weber also showed that the level of aggregation
not be understood as the final drivers of behaviour, but as factors is an important parameter in such decompositions, with fewer
that enable or prohibit people from engaging in various practices consumption categories leading to underestimation of the impor-
that then result in environmental pressure. The paper also aims to tance of structure.
improve the understanding of how results from cross-sectional In the following section, methods and data used in the paper
analyses relate to development over time. Cross-sectional ana- will be presented. The main results are presented in Section 4 with
lyses are practical in terms of data availability but we are often a sensitivity analysis of import assumptions in Section 5. Finally, a
more interested in processes of change, e.g. what happens to con- discussion with concluding remarks is given in Section 6.
sumption patterns as incomes rise? In the second part of the study a
time-series analysis of GHG emissions from Swedish households 2. Method
between 1993 and 2006 is performed. The trend is decomposed
into technical and structural components in order to enable com- Two methods are used to address the research question of this
parison to the cross-sectional analysis. paper; a multivariate regression model (Section 2.2) and a
More specifically, the paper addresses the following research decomposition analysis of trends (Section 2.3).
questions:
2.1. Indicators of climate impact from consumption
What are the relationships between consumption, energy use
and GHG emissions for Swedish households? While the context of this paper is climate change mitigation,
What other determinants explain the variance in energy use and primary energy is used as an indicator in conjunction with CO2
GHG? equivalents. This is due to the aforementioned decarbonisation of
How do these cross-sectional estimates relate to development some sectors of the Swedish economy. These changes depend
over time? mainly on the dynamics of the supply side and have very little to do
with changing consumption patterns. Since bioenergy currently
The first two of these questions have been addressed in previous dominates the Swedish heat sector, consumption of heat gives rise
research although not with this data set. Several cross-sectional to very low CO2 emissions, but from an energy systems perspective
studies have shown that energy demand and GHG emissions consumption of heat also binds bioenergy resources that could
calculated from the consumption side are strongly dependent on have been used to replace fossil fuels in other sectors. Likewise,
income levels although the relationship is typically somewhat while the Swedish electricity mix is low in CO2 intensity, increas-
lower than 1:1. National multivariate regressions using a set of ingly interconnected power systems mean that reduced demand of
socio-economic parameters in Lenzen et al. [11] showed that direct electricity in Sweden would enable reductions of emissions in the
and indirect energy use of households increase sharply with in- European system as a whole. Straining these arguments, the most
come also in affluent countries, with expenditure elasticities of 0.64 climate friendly private consumption pattern may rather be that
in Japan, 0.78 in Australia, 0.86 in Denmark and India, and 1.00 in which requires the least energy. However a main drawback of
Brazil. Moreover, for all of these countries except India, a larger studying only primary energy is that it does not capture the link
degree of urbanity was related to lower energy use. Shammin et al. between consumption and non-energy GHGs like CH4 and N2O
[12] estimated the expenditure elasticity of energy in the US to be from food production. Hence both indicators are used throughout
0.57. This study also found that parameters related to urban sprawl the paper.
play significant roles in determining the energy intensity of
households. Estimates of expenditure elasticities for GHG emis- 2.2. Multivariate linear regression analysis
sions are in the same range as for energy; e.g. 0.84 in the
Netherlands [13] and 0.91 in Spain [14]. Hertwich and Peters [15] The cross-sectional part of the study is based on multivariate
found similar results also from a cross-national comparison with linear regression (OLS) modelling of energy use and GHG emissions
an expenditure elasticity of GHG emissions of 0.57 (0.81 for CO2 and from Swedish households according to Eq. (1).
0.32 for the other gases). This result is based on an analysis using a
multi-regional inputeoutput model with trade statistics for 57 ln E ¼ C þ a ln Y þ b ln H þ h1 D1 þ h2 D2 þ / þ hn Dn (1)
sectors and 73 nations, but with expenditures as the only explan-
atory variable. where E is energy use (direct and indirect), Y is total annual con-
Time-series analyses on GHG emissions are fewer in the litera- sumption and H is household size and D1 to Dn is a set of dummy
ture. There are a fair number of studies addressing the relative variables. The same model is also used replacing energy use with
importance of intensity and structure on the production side of the GHG emissions as the dependent variable. All variables are
economy (see e.g. Refs. [16e19]), while only a few approach this described in Table 1. The regression coefficient a is the elasticity of
from the consumption side. Druckman and Jackson [20] studied the energy use with respect to total consumption. If Y increases by 1
carbon footprint from UK households during a similar time period percent, E increases by a percent. The dummy variables can be
as in this paper (1990e2004), and found that CO2 emissions interpreted so that when a dummy D is 1, then the exponential
decreased during the first half of the nineties as a result of fuel function of its regression coefficient (eh) is a multiplier for E in
substitutions in the electricity sector, but that emissions have since relation to the reference case (when D is 0).
been on the rise, mainly due to increasing CO2 emissions The data used to estimate the parameters in Table 1 are
‘embedded’ in products and services. Munksgaard et al. [21] stud- described in Section 2.3.
ied energy use related to Danish household consumption between
1966 and 1992 and showed that the effect of growth in total con- 2.3. Decomposition analysis
sumption was mainly counteracted by decreasing energy in-
tensities in the production sectors, while the changing structure of One of the questions raised in this paper is how estimates based
consumption played only a minor role. Weber [22], however, ana- on differences between households (cross-sectional analysis) relate
lysed a shorter time period for the US (1997e2002) and found that to development over time. Since the development of GHG
100 J. Nässén / Energy 66 (2014) 98e106
Table 1 affects the absolute value of the terms DUi but not their relative
Description of dependent and independent variables used in the regression models. importance in the decomposition, which is the focus of this paper.
Dependent variables Type Description In this study, five factors Bt,i (Eq. (2)) have been identified. The
Energy use log Total annual primary energy usea
specific expression of Ut is given in Eq. (5).
per household
Including direct and indirect energy use Ut Et St Ct
Ut h $ $ $ $Pt (5)
of products and services Et St Ct Pt
GHG emissions log Total annual CO2 equivalent emissions
(CO2, CH4, N2O) per household where E is primary energy, S energy service demand (defined as
Including emissions from energy use,
agriculture and industrial processes,
energy use with constant technical energy efficiency), C total con-
but not from deforestation sumption in fixed prices, and P population. Each of these variables,
except P is a sum over 104 categories of products and services j
Independent Type Description
variables (listed in the Appendix):
primary energy use and CO2 equivalents (global warming potential, are divided by the size of the households measured in consumption
GWP over 100 years for carbon dioxide, methane and nitrous oxide) units (c.u.), since otherwise the plot would to a large extent only
per unit of final consumption are calculated using monetary describe the difference between large and small households, e.g.
transactions between sectors together with multipliers of direct single income households in the lower left and double income
energy use and emissions in each sector. Thus the method re- households in the upper right (see footnote b of Table 1 for a
allocates energy use and emissions from production sectors description of the c.u. measure). The exponent of the fitted power
where they occur to the final consumption of goods and services, function 0.78 is a simple estimate of the elasticity of energy use
including indirect contributions from an unlimited number of up- with respect to total expenditures. The corresponding elasticity for
stream production activities. GHG emissions is 0.83.
The Mirdata database contains only primary inputs of fossil fuels Although the simple representation of Fig. 1 gives a surprisingly
and bioenergy and final demand of electricity and district heating high coefficient of determination for only one explanatory variable
[28]. Inputs of nuclear, hydro and wind power have been added in (an R2 of 0.55), there is still a significant variance within each level
accordance to their respective shares of electricity generation in of consumption. Households with annual expenditures of 20,000 V
Sweden. Primary energy for electricity has been calculated with per c.u. may for example cause energy demand from around 100 to
efficiencies of 0.37 for nuclear and thermal power. For hydro and over 300 GJ per c.u. and year.
wind, primary energy is calculated as produced electricity plus Fig. 2 shows frequency distributions for households’ with
internal energy use. This gives a total weighted average conversion different energy and GHG intensities. The distribution for energy
efficiency of 0.52 from primary energy to electricity for the studied intensity (left diagram) is rather wide and dividing the set into
period. households living in single-family (detached or row) houses and
In the Mirdata energy use and GHG emissions for imported households living in apartment buildings also shows a clear divi-
goods and services are calculated as if they had been produced sion with much lower energy intensities for the apartments. This
domestically. Since the Swedish energy system, particularly the effect is visible also for GHG intensities (right diagram) but not as
electricity sector, is relatively low in CO2 emissions, this results in strong as for energy intensity. This is due to that the main difference
an underestimation of the emissions for imported goods [29]. These between households in apartments and detached buildings is the
errors may be significant in absolute terms, but less important for demand for electricity and heat. Particularly households living in
comparisons between households or for relative changes over time. single-family houses with electric heating cause a very high de-
The sensitivity of the results with respect to this assumption is mand of primary energy, but since the Swedish electricity mix is so
tested in Section 5. low in carbon emissions this does not add much to the GHG in-
tensity. The main reason for the remaining difference in GHG in-
tensity is that households in detached buildings also spend a larger
4. Results
share on transport fuels.
Figs. 1 and 2 show that income/total expenditures and type of
Results from the cross-sectional (CS) regressions and time-
dwelling are two important variables, but to test a larger set of
series (TS) decomposition are presented in Sections 4.1 and 4.2,
variables we need a formal multivariate analysis. Table 2 shows the
respectively, and compared in Section 4.3.
results of a regression analysis based on Eq. (1) with variables
described in Table 1.
4.1. Cross-sectional analysis As shown in Table 2, increasing total consumption by 1 percent
corresponds to increases in energy use by 0.77 percent and GHG
This part of the study is primarily based on a multivariate emissions by 0.85 percent. This is very similar to the elasticities
regression (OLS) analysis performed according to the model found in the univariate regression (Fig. 1). As was also shown in
described in Section 2.2. But, to get a first visual description of the Fig. 2, the type of housing is also a very important determinant. In
data, a scatter plot of energy use as a function of consumption (total addition to this difference, rural households also seem to require
annual expenditures) is presented in Fig. 1. The data on both axes around 8% more energy than urban households, but this effect is
more ambiguous and statistically not significant for GHG emission.
The effects of all age groups are statistically significant, with higher
energy and GHG intensities for older age groups. Education has a
large influence on some specific categories of consumptions but for
aggregated energy use and GHG emissions the coefficients are
small and not statistically significant.
The independent variables of the regressions have been tested
for multicollinearity by means of Variance Inflation Factors (VIF).
The VIFs are low for all variables with the highest value 2.17 for
household size.
Fig. 2. Histograms of primary energy per V expenditures (left diagrams) and GHG emissions per V expenditures (right diagrams). The diagrams compare the frequency distri-
butions for households living in single-family (detached or row) houses and households living in apartment buildings.
from other non-durable goods. Durable and semi-durable goods consumption is found to be the main driver towards increasing
have been lumped into one category since these products have emissions while improvements in energy efficiency in production
similar energy intensities (jointly termed durable goods in Fig. 3). and end-use is the most important mitigating factor.
The food category also includes restaurant services. A complete list In this paper, however, we are particularly interested in the
of these categories is provided in the Appendix. structural component of energy intensity, i.e. the effect of the
The total consumption in fixed prices increased by 39 percent change in composition of consumption illustrated by Fig. 3. The
during the period. Consumption of durable goods such as vehicles, contribution of structural change to the total reduction in energy
appliances and furniture doubled during the period and accounted demand is less than half of the contribution from energy efficiency.
for almost half of the total increase (this category also includes The decomposition shows that an increase in consumption per
semi-durables, e.g. clothes and sport equipment). Services capita by 1% would have caused an increase in energy service de-
(excluding travel services) also increased significantly and mand (energy demand if energy efficiency would have been kept
accounted for around a quarter of the total increase. Expenditures constant) by 0.72%. Potentially, part of the development of con-
on food and housing increased slightly while direct energy, non- sumption patterns over time may also be driven by changes in
durable goods (consumables) and travel services remain close to relative prices. To test this a series of very simple regression ana-
constant. lyses was carried out with expenditures on each of the 104 con-
Fig. 4 shows the aggregated effect of these consumption changes sumption categories as dependent variables and total expenditures
on GHG emissions between 1993 and 2006 but decomposed into a and price index as the dependent variables. Adding price in this
series of factors according to the method described in Section 2.3. way changed the estimate of expenditure elasticity for energy
Each bar can be understood as the isolated contribution of that service demand from 0.72 to 0.80, still very close to the finding of
factor to the total change in emissions. Dark red (colour in the web the CS-analysis.
version) bars indicate an isolated effect towards increasing emis- The reduction in aggregated energy intensity due to structural
sions while light green (colour in the web version) bars indicate an change of the consumption is dominated by the same consumption
isolated effect towards decreasing emissions. Increasing total categories that increase the most, i.e. durable goods and services
Table 2
Results from the multivariate OLS regressions with energy use and CO2 equivalent emissions as the dependent variables. Regression coefficients presented together with
significance levels and heteroskedasticity-robust standard errors (rob. s.e.). The right column gives examples of consumption categories that are strongly and statistically
significantly affected by each of the independent variables. See Table 1 for a description of the included variables.
Energy use (log) CO2 eq. (log) Ex. of consumption categories that increase
with each parameter
Coeff. Rob. s.e. Coeff. Rob. s.e.
Total consumption (log) 0.769*** 0.018 0.852*** 0.015 Cars, electronics, clothes, package holidays
Household size (log) 0.151*** 0.029 0.148*** 0.021 Food, daycare services
Age under 301 0.140*** 0.021 -0.108*** 0.016 Rail travel, telephones, books
Age 30e451 -0.069*** 0.015 -0.073*** 0.011 Daycare services, games, hobbies
Age over 601 0.074*** 0.019 0.042** 0.015 Summer houses, pharmaceuticals
High education2 -0.016 0.013 0.023* 0.011 Hotels, rail travel, books
Low education2 0.009 0.017 0.019 0.013 Tobacco, papers & magazines
Single-family house3 0.418*** 0.013 0.092*** 0.011 Heating fuels, electricity
Rural4 0.057*** 0.014 -0.001 0.010 Vehicle parts, domestic air travel
Urban4 -0.021 0.014 0.027* 0.011 Public travel, package holidays, restaurants
Constant 0.939*** 0.094 2.776*** 0.077
N 1969 1969
Adj. R2 0.839 0.841
Reference categories: 1age 46e60; 2average education; 3apartement; 4small/medium size town.
Significance levels: *p < 0.05; **p < 0.01; ***p < 0.001 (two-tailed test).
J. Nässén / Energy 66 (2014) 98e106 103
Table 3
Comparison of expenditure elasticities between the CS and TS analyses for total
demand and for seven consumption categories (see Appendix for a list of what these
include). To make the results comparable, the TS results are calculated for energy
service demand, i.e. deducting the effect of improved technical energy efficiency
from the energy trend.
Expenditure elasticities
4.3. Comparison of results from cross-sectional and time-series 5. Sensitivity analysis: imported goods
analyses
The energy and GHG intensities used in the CS and TS analyses
Comparing the results from Sections 4.1 and 4.2 we see that of this paper are taken from the Mirdata database (Section 3.2) in
when the effect of improved energy efficiency is deducted from the which imported goods are assumed to be produced with the same
time trend in energy use, these analyses provide rather similar technology as domestic production. Since the Swedish energy
results at the aggregated level. The expenditure elasticities were system, particularly the electricity sector is comparatively low in
estimated to be 0.72 for energy use in the CS-analysis and 0.77 for CO2 emissions this causes an underestimation of the emissions for
energy service demand in the TS-analysis. In Table 3 this compar- imported goods. If the aim of this paper would have been to
ison is also done for the seven main categories of consumption calculate the total emissions related to Swedish consumption this
shown in Fig. 3. A list of the composition of these categories is given assumption would have caused significant errors. However, for
in the Appendix. relative comparisons of emissions between households or for the
From Table 3 we see that the results are rather similar for the relative development over time these errors can be expected to be
categories direct energy use, housing and non-durable goods. much smaller. In this section we use results from previous multi-
However, the demand for services and durable goods increased at regional inputeoutput analyses to quantify the effect of this
significantly higher rates over time (TS) than from households with assumption on the main results of the paper.
low to high levels of consumption (CS). The durable goods category Results from studies that take into account differences in pro-
has the highest expenditure elasticity in both analyses, but this duction technology between different countries vary considerably
effect is even more pronounced in the TS-analysis. Looking at this due to differences in assumptions and models. A report from SEPA
category more in detail, we can see that the consumption of goods [9] found that the Swedish GHG emissions are 25% higher if
related to for example telecommunication and computers grow consumption-based accounting is used instead of traditional
production-based accounting. This corresponds to a GHG intensity
1.4 of imports which is 1.5 times as high as the domestic production.
The results from Peters and Hertwich [31] imply a higher ratio of 1.9
studying only CO2 emissions. Using figures from Davis and Caldeira
GHG em issions (index 1993 = 1)
[10] gives an even higher ratio of 2.9 also only accounting for CO2
1.2
emissions. A difference between these studies is that while Peters
and Hertwich only account for bilateral trade, the model used by
Davis and Caldeira also includes trade in several steps, e.g. emis-
U1993 sions from production in China that is exported to intermediate
1.0
U1993
U2006 U2006
production in Germany and then further exported for final con-
sumption in Sweden are accounted for as Swedish emissions.
A first question is whether these differences are due to differ-
0.8 ences in the structure of imports in relation to domestic production
Population Consumption Structure of Energy Emissions per Total change of
per cap consumption efficiency energy use emissions or if they are due to differences in the CO2 intensity of each con-
sumption category.3 The importance of the structural component
Fig. 4. Decomposition of the change in GHG emissions from Swedish household
can be tested by comparing the aggregated CO2 intensities of
consumption between 1993 and 2006 into underlying factors. The decomposition is
performed according to the additive Parametric Divisia Method (Section 2.3). Each bar
can be understood as the isolated influence of that factor, given that all other factor
3
would have been held constant. Dark red bars indicate an isolated effect towards Here we interpret this primarily as technological differences such as fuel
increasing emissions while light green bars indicate an isolated effect towards choices and energy efficiency. However it should be noted that CO2 intensities of
decreasing emissions. [For interpretation of colour referred in this figure legend, the imported goods from countries with lower costs of labour are also higher since this
reader is referred to web version of the article.] causes fuel to make up a larger share of the total cost.
104 J. Nässén / Energy 66 (2014) 98e106
domestic production and imports in the Mirdata database [28], proportional increase in energy use and emissions, since part of the
where CO2 intensities are assumed to be identical for each of the increase in consumption may be regarded as improved quality
104 consumption categories. Excluding the emissions from the rather than quantity. Since quality is more coupled to input of la-
households’ direct energy use we find that the aggregated CO2 in- bour than to input of materials, it is likely that it also has a weaker
tensity is actually 4% higher for domestic production than for im- relationship to energy use and emissions than increased quantity.
ports. Hence the structural component is very small in relation to Girod and de Haan [32] addressed this issue by constructing a
the total difference found by SEPA [9], Peters and Hertwich [31] and model of households’ GHG emissions which is based on functional
Davis and Caldeira [10] and it also goes in the other direction. Based units (e.g. kg of food and clothes) instead of the monetary units (e.g.
on this we continue to test the sensitivity of the results with respect V of food and clothes). Using Swiss consumption data coupled to an
to import assumptions by adjusting the CO2 intensities themselves LCA database, this approach reduced the elasticity of GHG emis-
disregarding the structural difference of imports and domestic sions with expenditure from 0.94 to 0.53. So which of these esti-
production. mates is more correct? Vringer and Blok [33] showed that at least
Re-running the regression with adjusted emissions intensities for clothes and shoes the actual dependence of energy on price lays
for imported goods results in an increase in the expenditure elas- in-between estimates based on monetary and functional units.
ticity of GHG emissions from 0.85 (main assumptions, Table 2) to Hence, while monetary units may result in overestimates of the
0.89 (factor 1.5 [9]). 0.92 (factor 1.9 [31]) or 0.97 (factor 2.9 [10]). expenditure elasticity of energy and GHG emissions, functional
Hence higher emissions intensities of imports give an even stronger units may result in underestimates. This methodological issue
coupling between consumption and emissions. This can be warrants more research.
explained by the larger share of imported products in the expen- We conclude that the relationship between consumption, en-
ditures by high-income households. The other regression co- ergy use and emissions is very strong and that the variable total
efficients remain almost unaffected. expenditures explain more of the variation between households
Similarly, since the share of imported goods also increases over than any other variable.
time; higher GHG intensities of imports also result in a weaker
decoupling between emissions and consumption in the time-series 6.2. What other determinants explain the variance in energy use
analysis. Using the main assumptions, GHG emissions was found to and GHG?
decrease slightly by 4% between 1993 and 2006 (Fig. 4). With
inflated emissions intensities for imports, the total GHG emissions Total expenditures aside, this study indicates that spatial de-
are instead found to increase during the same period, by 2% (factor terminants such as where a household decides to live play an
1.5 [9]), 6% (factor 1.9 [31]) or 15% (factor 2.9 [10]). However, the important role for the overall energy use and emissions. Particu-
difference between imported and domestically produced goods is larly the lifestyles of households living in single-family houses
much smaller in terms of energy use than for CO2 emissions. While where found to cause much higher total energy demand than
the results by Davis and Caldeira [10] show that Swedish imported households living in apartments. The detail in which these factors
goods have CO2 intensities that are 2.9 times as high as the do- have been analysed, however, is highly constrained by the avail-
mestic production, the difference is only 18% for energy use. Hence ability of data. The only parameters tested being dummy variables
it can be concluded that the technological difference is primarily representing single-family house/apartment and rural/small to
found in the factor to the right of Fig. 4 (emissions per energy use) medium town/large city. Hence potentially important aspects such
while the results for energy efficiency are more robust. as commuting distances and differences due to for example city
density have not been analyzed and require other data sets.
6. Discussion and conclusions Everything else equal, age also showed a surprisingly strong
positive relationship with both total energy use and GHG emis-
In this final section, the results of the analysis (Sections 4 and 5) sions. Main factors behind this are increasing shares of expendi-
are discussed in relation to the three questions raised in the tures on residential energy with age and a relatively low share of
introduction. expenditures on transport fuels in the youngest age group. The
level of education on the other hand did have a significant impact
6.1. What are the relationships between consumption, energy use on the relative expenditure shares, but not on the aggregated en-
and GHG emissions for Swedish households? ergy use and GHG emissions.
While motivational determinants of consumption were not part
In the cross-sectional analysis presented in Section 4.1, the of the analysis, the results can at least be discussed and related to
expenditure elasticities were estimated to be 0.77 for primary en- theoretical concepts. One such point is the importance of status and
ergy use and 0.85 for GHG emissions. This means that there is a very identity in shaping consumption patterns (conspicuous consump-
strong link between total expenditures and emissions. On average, tion [34]; positional consumption [35]) in relation to emissions that
a household with twice as high expenditures as another household relate to habitual behaviours (ordinary consumption [36]). The
also causes 80% more GHG emissions. These findings fall within the results from this study indicate that even if the bulk of emissions is
range of previous studies on households in other countries which related to habits, for example diet, space and water heating, and
have found expenditure elasticities from 0.57 in the US to 1.00 in energy for daily travel such as commuting, consumption for status
Brazil [11e14]. or identity may be more important on the margin. An example of
These results are based on a method where imported goods are this is expenditures on cars which have has been shown to be a
assumed to be produced with the same technology (GHG intensity) highly positional good [37] and which account for over 20% of the
as domestic production. Accounting for differences in the GHG in- marginal expenditures (see also Holmberg and Nässén [38]).
tensity of imports the expenditure elasticity of GHG emissions in- Hence, while positional consumption account for a small share of
crease further towards unity, to between 0.89 and 0.97 depending emissions it may constitute an important driver towards increasing
on the assumptions used (see Section 5). total consumption which is also found to be the main factor to-
However another potential error source may also cause over- wards increasing emissions in the decomposition (Fig. 4).
estimations of these elasticities. When high-income households Further surveys ought to integrate data collection on
spend more money on for example clothes this may not give a consumption-related GHG emissions with both motivational
J. Nässén / Energy 66 (2014) 98e106 105
determinants (attitudes, norms etc.) and structural determinants series than in the cross-sectional estimate. Particularly sub-
(urban form, availability of public travel etc) to enable interdisci- categories related to telecommunication and computers grew
plinary research to further the understanding of the variance of much faster over time. Such rapid increases are likely to be related
GHG emissions from households. to technological change in these sectors which also created entirely
new products.
6.3. How do these cross-sectional estimates relate to development The findings of this paper provide some support for the use of
over time? cross-sectional estimates in scenarios of aggregated energy de-
mand, but such scenarios would also need to be supplemented by
The energy intensity of Swedish household consumption separate scenarios of technical energy efficiency in both production
decreased significantly by around 2.3 percent per year during the sectors and end-use. Moreover, for long-term scenarios it is
studied period. About a third of this or 0.7 percent per year can be important also to recognize that consumption patterns may also be
attributed to the structural change of the composition of con- affected by the emergence of new technologies and products. Such
sumption. In this paper we have used the term energy service de- shifts are inherently difficult or impossible to handle with empirical
mand to describe an imaginary development of energy demand if methods.
technical energy efficiency would have remained constant, i.e.
changes in energy service demand occur only as a result of changes Acknowledgements
in the total amount and composition of consumption. In Section 4.2
the expenditure elasticity of energy service demand was calculated Funding from E.On and the Swedish Energy Agency (AES pro-
to be 0.72e0.80. This result is very similar to the expenditure elas- gramme) is gratefully acknowledged. Thanks to Daniel Johansson,
ticity of energy use of 0.77 found in the cross-sectional analysis Jörgen Larsson and John Holmberg for valuable comments on the
(Section 4.1), which is also only affected by differences in the manuscript.
composition of consumption at different levels of total expenditures.
A further comparison between these analyses showed that the
estimates were similar also for the sub-categories direct energy use, Appendix. Classification of consumption categories
housing and non-durable goods (Section 4.3). The resemblance in
the category of direct energy use, which has the highest energy The table provides the classification of final expenditures in 104
intensity, is also the main reason behind the similarity at the categories used in the paper. The seven groups are those used in
aggregated level. However, for specific types of goods the resem- Fig. 3. See Eurostat-OECD [30] for further description of this clas-
blance is not as strong. For example, the expenditure elasticity for sification. Energy and GHG emissions data for these categories are
durable goods was found to be significantly higher in the time- available from: http://www.mirdata.scb.se/.
Table A1
104 Consumption categories divided into seven groups together with COICOP code.
Table A1 (continued )
054 Glassware, tableware and household utensils 0723 Maintenance and repair of vehicles
0551 Major tools and equipment 07241-4 Parking, driving lessons, vehicle tests etc
0552 Small tools and accessories 0811 Postal services
0613 Therapeutic equipment 0813 Telephone and telefax services
0711 Cars 0915 Repair of audio-visual, photogr., info. equipment
0712 Motor cycles 0923 Repair of major durables for recreation
0713 Bicycles 0935 Veterinary services
0721 Spare parts for vehicles 0941 Recreational and sporting services
0812 Telephone and telefax equipment 0942 Cultural services
0911 TV sets, radios, gramophones etc 0943 Games of chance
0912 Photographic and cinematographic equipment 10 Education
0913 Information processing equipment 112 Accommodation services
0914 Recording media 1211 Hairdressing, personal grooming
0921 Major durables for outdoor recreation 12401 Daycare, kindergarten
0922 Music instruments, durables for indoor recreation 12402 Old age care
0931 Games, toys and hobbies 12403 Personal assistant
0932 Equipment for sport, and outdoor recreation 12404 General care
0951 Books 125 Insurance
1212 Electric appliances for personal care 126 Financial services
1231 Jewellery, watches 127 Other services
1232 Other personal effects
References [19] Andreoni V, Galmarini S. Decoupling economic growth from carbon dioxide
emissions: a decomposition analysis of Italian energy consumption. Energy
2012;44(1):682e91.
[1] Heston A, Summers R, Aten B. Penn world table version 7.1. Center for In-
[20] Druckman A, Jackson T. The carbon footprint of UK households 1990e2004: a
ternational Comparisons of Production, Income and Prices at the University of
socio-economically disaggregated, quasi-multi-regional input-output model.
Pennsylvania; July 2012.
Ecol Econ 2009;68(7):2066e77.
[2] Olivier JGJ, Janssens-Maenhout G, Peters JAHW, Wilson J. Long-term trend in
[21] Munksgaard J, Pedersen KA, Wien M. Impact of household consumption on
global CO2 emissions e 2011 report. The Hague: PBL/JRC; 2011. www.pbl.nl/
CO2 emissions. Energy Econ 2000;22(4):423e40.
en/publications/2011/long-term-trend-in-global-co2-emissions-2011-report.
[22] Weber CL. Measuring structural change and energy use: decomposition of the
[3] Rogelj J, Hare W, Lowe J, van Vuuren DP, Riahi K, Matthews B, et al. Global
US economy from 1997 to 2002. Energy Policy 2009;37(4):1561e70.
emission pathways consistent with a 2 C global temperature limit. Nat Clim
[23] Statens Offentliga Utredningar. Förbättrad statistik om hushållens inkomster.
Change 2011;1:413e8.
Report No. SOU 2002:73. Stockholm: Justitiedepartementet; 2002 [Swedish].
[4] Meadows DH, Randers J, Meadows DL. The limits to growth: the 30-year
www.regeringen.se/sb/d/108/a/462.
update. USA: Chelsea Green; 2004.
[24] Ang BW. Decomposition methodology in industrial energy demand analysis.
[5] Jackson T. Prosperity without growth: economics for a finite planet. New
Energy 1995;20(11):1081e95.
York: Earthscan; 2009.
[25] Trafikanalys. Körsträckor. Swedish. Available from: http://www.trafa.se/sv/
[6] Azar C, Schneider SH. Are the economic costs of stabilising the atmosphere
Statistik/Vagtrafik/Korstrackor; 2008.
prohibitive? Ecol Econ 2002;42(1e2):73e80.
[26] Swedish Energy Agency. Energy statistics for dwellings and non-residential
[7] van den Bergh JCJM. Safe climate policy is affordable-12 reasons. Clim Change
premises 2008. Report No. ES2009:10. Eskilstuna; 2009. http://webbshop.cm.
2010;101(3):339e85.
se/System/TemplateView.aspx?p¼Energimyndigheten&view¼default&cat¼%
[8] Kander A. Economic growth, energy consumption and CO2 emissions in
2FRapporter&id¼2bdad6589c6a4dabb6edb7609fb1d305.
Sweden 1800e2000 [dissertation]. Lund studies in economic history 19. Lund
[27] Statistics Sweden. Household budget survey (HBS) 2005e2007, expenditure
University; 2002.
and income report. Report No. HE 35 SM 0801. Örebro; 2008. www.scb.se/
[9] SEPA. Konsumtionens klimatpåverkan. Report No. 5903. Stockholm: Envi-
statistik/HE/HE0201/2005I07/HE0201_2005I07_SM_HE35SM0801.pdf.
ronmental Protection Agency; 2008 [Swedish]. www.naturvardsverket.se/
[28] Statistics Sweden. Environmental accounts. Database. Available from: www.
Om-Naturvardsverket/Publikationer/ISBN/5900/978-91-620-5903-3/.
mirdata.scb.se.
[10] Davis SJ, Caldeira K. Consumption-based accounting of CO2 emissions. Proc
[29] Carlsson-Kanyama A, Assefa G, Peters G, Wadeskog A. Koldioxidutsläpp till
Natl Acad Sci U S A 2010;107(12):5687e92.
följd av Sveriges import och konsumtion: beräkningar med olika metoder.
[11] Lenzen M, Wier M, Cohen C, Hayami H, Pachauri S, Schaeffer R. A comparative
TRITA-IM:11. Stockholm, Swedish; 2007.
multivariate analysis of household energy requirements in Australia, Brazil,
[30] Eurostat-OECD. Methodological manual on purchasing power parities. Paris:
Denmark, India and Japan. Energy 2006;31(2e3):181e207.
European Commission; 2006.
[12] Shammin MR, Herendeen RA, Hanson MJ, Wilson EJH. A multivariate analysis
[31] Peters GP, Hertwich EG. CO2 embodied in international trade with implica-
of the energy intensity of sprawl versus compact living in the U.S. for 2003.
tions for global climate policy. Environ Sci Technol 2008;42(5):1401e7.
Ecol Econ 2010;69(12):2363e73.
[32] Girod B, de Haan P. More or better? A model for changes in household
[13] Kerkhof AC, Nonhebel S, Moll HC. Relating the environmental impact of
greenhouse gas emissions due to higher income. J Ind Ecol 2010;14(1):31e49.
consumption to household expenditures: an inputeoutput analysis. Ecol Econ
[33] Vringer K, Blok K. Assignment of the energy requirements of retail trade to
2009;68(4):1160e70.
products. The Netherlands: Department of Science, Technology and Society,
[14] Roca J, Serrano M. Income growth and atmospheric pollution in Spain: an
Utrecht University (STS-UU); 1996.
inputeoutput approach. Ecol Econ 2007;63(1):230e42.
[34] Veblen T. The theory of the leisure class. New York: Mcmillan; 1899.
[15] Hertwich EG, Peters GP. Carbon footprint of nations: a global, trade-linked
[35] Hirsch F. Social limits to growth. Cambridge: Harvard University Press; 1976.
analysis. Environ Sci Technol 2009;43(16):6414e20.
[36] Groncow J, Warde A. Ordinary consumption. London: Routledge; 2001.
[16] Ang BW, Liu XQ, Ong HL. Sector disaggregation and the effect of structural
[37] Alpizar F, Carlsson F, Johansson-Stenman O. How much do we care about
change on industrial energy consumption. Energy 1992;17(7):679e87.
^lu G. A sectoral decomposition analysis of Turkish CO2 emissions absolute versus relative income and consumption? J Econ Behav Organ
[17] Kumbarog
2005;56(3):405e21.
over 1990e2007. Energy 2011;36(5):2419e33.
[38] Holmberg J, Nässén J. Well-being the path out of the consumption-climate
[18] Hammond GP, Norman JB. Decomposition analysis of energy-related carbon
dilemma? In: Ekström K, Glans K, editors. Beyond the consumption bubble.
emissions from UK manufacturing. Energy 2012;41(1):220e7.
New York: Routledge; 2011.